The Agent Skills Directory

[PROMPT_INJECTION]: The skill has an indirect prompt injection attack surface because it processes untrusted data from external trademark databases and user-provided descriptions, then writes generated reports to the local file system. While the skill employs structural templates and mandatory safety disclaimers, it lacks technical sanitization for content retrieved from external tools.
Ingestion points: Untrusted data enters the agent context through user descriptions of proposed trademarks and search results retrieved from the CNIPA trademark database or other connected tools.
Boundary markers: The skill mandates the use of specific markdown headers and a standardized disclaimer block in all outputs to separate generated content from system instructions, but does not use machine-enforced delimiters for external data.
Capability inventory: The skill possesses the capability to read configuration files and write memorandum documents to specific matter-related directories within the user's home directory (~/.claude/plugins/config/claude-for-legal/ip-legal/).
Sanitization: There is no evidence of filtering, escaping, or strict schema validation for the data retrieved from external trademark databases prior to its interpolation into the final report.

clearance